Patient characteristics
From December 4, 2020, to August 4, 2023, 36 patients were enrolled and all received camrelizumab plus apatinib and SOX. Of these, 10 patients discontinued combination treatment due to disease progression (n = 6), withdrawal of consent (n = 3), and adverse events (n = 1). Consequently, 26 patients (72.2%) continued to receive camrelizumab plus apatinib as maintenance treatment (Fig. 1). Antitumor activity and safety were analyzed in all 36 patients. The data cutoff for safety and efficacy analysis was March 27, 2024. Following the end of study treatment, 21 patients (58.3%) received at least one subsequent antitumor treatment (supplementary Table 1).
The median age was 63 years (range, 28-78), and 30 patients (83.3%) were male. Most patients had gastric cancer (n = 23, 63.9%) and were in stage IVB (n = 31, 86.1%). In addition, 18 patients (50.0%) had liver metastasis, 17 (47.2%) had distant lymph node metastasis, and 9 (25.0%) had peritoneal metastasis. Median serum AFP level was 739.8 ng/ml (range: 77.7-321,847.0). Programmed cell death ligand 1 (PD-L1) expression was observed as CPS ≥ 1 in 18 patients (50.0%), CPS ≥ 5 in 11 patients (30.6%), and CPS ≥ 10 in 5 patients (13.9%). The baseline characteristics were summarized in Table 1.
Efficacy
In the full analysis set, the overall confirmed ORR per RECIST v1.1 was 66.7% (95% CI: 49.0-81.4). The disease control rate (DCR) was 88.9% (95% CI: 73.9-96.9). Two patients had complete response (CR), 22 had partial response (PR), and 8 had stable disease (SD) (Table 2). The maximal tumor shrinkage of target lesion relative to baseline is depicted in a waterfall plot (Fig. 2a). Notably, all SD patients and one PD patient had tumor shrinkage from baseline, with one patient with unconfirmed PR (recorded as SD) showing 55.3% shrinkage. Treatment durations and tumor response are shown in swimmer and spider plots (Fig. 2b, c). The median time to response (TTR) was 1.4 months (95% CI: 1.4-2.0). The median duration of response (DoR) was 8.4 months (95% CI: 3.9-not reached [NR]).
Antitumor response and survival outcomes. a Waterfall plot showing the percentage change in the sum of target lesion diameters from baseline. b Swimmer plot showing treatment exposure and response duration. Patients 1004 and 1047 had partial response, and subsequently underwent surgery. Both patients had R0 resection, with NCCN tumor regression grades of 1 and 3, respectively. As of the data cut-off date, both patients had no recurrence or death. c Spider plot showing the percentage change in the sum of target lesion diameters during treatment. Tumor response per RECIST v1.1 was evaluable in all patients. d Kaplan-Meier curve of progression-free survival. e Kaplan-Meier curve of overall survival
During a median follow-up of 11.7 months (range: 3.2-37.9), 23 events (22 progressors and one death) occurred for PFS and 17 events for overall survival (OS). The median PFS was 7.8 months (95% CI: 4.9-12.3) (Fig. 2d). The 9-month and 12-month PFS rates were 39.2% (95% CI: 22.0-56.0) and 35.3% (95% CI: 18.7-52.3), respectively. The median OS was 18.0 months (95% CI: 10.5-NR) (Fig. 2e). The 9-month and 12-month OS rates were 76.7% (95% CI: 58.7-87.6) and 67.0% (95% CI: 48.2-80.3), respectively.
Subgroup analysis of ORR and PFS are shown in the supplementary Fig. 1 and supplementary Table 2. In patients with CPS ≥ 1, the ORR was 72.2%, compared to 71.4% in those with CPS < 1. In patients with CPS ≥ 5, the ORR was 81.8%, compared to 66.7% in those with CPS < 5. ORRs were similar across different AFP levels, with cutoffs at 400 or 800 ng/ml. Patients with GEJ tumor location, intestinal type, or with 1-2 metastatic organs showed a trend toward higher ORRs.
Safety
Any grade treatment-related adverse events (TRAEs) were observed in 34 patients (94.4%) (supplementary Table 3). And 16 patients (44.4%) had grade ≥3 TRAEs, with decreased neutrophil count (6 [16.7%]), hypertension (3 [8.3%]), anemia (2 [5.6%]), decreased platelet count (2 [5.6%]), and diarrhea (2 [5.6%]) being the most frequently occurring events. No treatment-related deaths occurred. Any grade immune-related adverse events (irAEs) were reported in 18 patients (50.0%). The most frequent irAEs were RCCEP, increased amylase level, and hyperthyroidism (supplementary Table 3). Grade ≥3 irAEs occurred in 3 patients (8.3%), including diarrhea, decreased platelet count, and increased lipase level (n = 1 each). A detailed summary of TRAEs and irAEs is available in the supplement. Treatment-related serious adverse events (SAEs) were reported in 3 patients (8.3%), including diarrhea (2 [5.6%]), vomiting (one [2.8%]) and hyponatremia (one [2.8%]).
No patient discontinued all study drugs due to TRAEs. Three (8.3%) patients discontinued any study drug: one patient discontinued chemotherapy due to decreased neutrophil, platelet, and white blood cell counts; one patient discontinued camrelizumab due to decreased platelet count; and one patient discontinued apatinib due to fatigue. Thirteen (36.1%) patients required dose interruption or delay due to TRAEs, and 8 (22.2%) patients required dose reduction due to TRAEs. Details of dose interruption, delay, or reduction are shown in the supplementary Table 4.
Exploratory outcomes
To identify potential biomarkers predictive of durable clinical response and further elucidate the molecular features of AFP-G/GEJ adenocarcinoma, targeted exome next-generation sequencing (NGS) and multiplex immunofluorescence (mIF) were performed (Fig. 3a). Patients with CR, PR, and SD lasting at least 6 months were classified in the durable clinical benefit (DCB) group. Patients with PD or PR and SD lasting less than 6 months were classified in the no durable benefit (NDB) group.
Exploratory study overview and genomic characteristic in study cohort. a Schematic summary of the design of exploratory analyses. b Gene sequencing of 25 patients with available pre-treatment tumor tissue from the study cohort
Genomic alterations
Gene sequencing with qualified tumor samples and matched normal tissues was performed in 25 patients (Fig. 3b). The most frequent disease-related somatic alterations were TP53 (80%), MCL1 (32%), CCNE1 (24%), and ZNF217 (24%). The frequency of TP53 mutation was significantly higher compared to both TCGA-STAD all-stage samples and TCGA-STAD advanced-stage samples only, while the frequency of mutation in ARID1A was significantly lower (supplementary Fig. 2a, b), which was consistent with previous studies.20 Further, we observed that LRP1B mutation and PI3K pathway alteration were only present in the NDB group (Fig. 3b; supplementary Fig. 3a, b), which revealed that LRP1B mutation and PI3K pathway alteration were significantly associated with NDB, i.e., poorer PFS. However, our cohort did not find any association between tumor mutation burden (TMB), copy number variants (CNV) burden, and clinical efficacy (supplementary Fig. 3c).
Association between the density of tumor-infiltrating cells and durable clinical benefit
We conducted a 4-panel mIF assay to provide a comprehensive and systematic presentation of the overall AFP-G/GEJ adenocarcinoma TME landscape (Fig. 4a; supplementary Fig. 4-7). The relationship between durable clinical efficacy and tumor-infiltrating cells (TICs) density was analyzed. We observed that patients with DCB had higher CD3+ T, CD4+ T and CD8+ T cells densities than patients with NDB (Fig. 4b). Further analyses revealed that immune checkpoints, including PD-1+ cells, FOXP3+ cells (supplementary Fig. 8a), and conventional immune checkpoint-positive T cells, such as PD-1+CD8+ T cells, PD-L1+CD4+ T cells, and CD4+FOXP3+CTLA4+ T cells, were significantly enriched in the DCB group (Fig. 4b). The cell density in the cohort was classified as “high” for densities falling within the top third of the group, and “low” for densities within the bottom third. A co-analysis of PD-1+ (CD8+) and FOXP3+ (CD4+) cells revealed that the proportion of patients exhibiting co-high expression of both FOXP3+ (CD4+) and PD-1+ (CD8+) cells, defined as (FOXP3+CD4+)hi(PD-1 + CD8 + )hi or FOXP3hiPD-1hi, was higher in those who demonstrated DCB compared to the non-co-high expression group (83.3% vs 45.0, p = 0.170). Additionally, when patients with low expression of both markers, categorized as (FOXP3+CD4+)low(PD-1+CD8+)low or FOXP3lowPD-1low, were analyzed together, the proportion of DCB was 28.6% in those with (FOXP3+CD4+)low(PD-1+CD8+)low (28.6% vs. 83.3%, p = 0.0079) and 21.4% in those with FOXP3lowPD-1low (21.4% vs. 91.7%, p = 0.0005) (supplementary Fig. 8b). Then, we compared the density of myeloid and stromal cells between the DCB and NDB groups, but no differences were observed (Fig. 4b). Tertiary lymphoid structures (TLSs) are key components of the TME,21,22 and we evaluated the predictive value of TLSs for ICIs-combined therapy in patients with AFP-G/GEJ adenocarcinoma. Compared with TLSs-negative patients, TLSs-positive patients had a better tendency for DCB (8 [66.7%] of 12 vs 6 [42.9%] of 14, p = 0.267) (supplementary Fig. 9). Additionally, TLSs were identified in 14 cases (46.6%), a rate considerably lower than the prevalence of TLSs observed in AFP-negative GC.23,24 These observations suggest that the TME of AFP-G/GEJ adenocarcinoma may be “colder” than that of AFP-negative gastric adenocarcinoma.
Overall TICs landscapes and comparison of TICs density between DCB and NDB groups. a TICs density grouped by subtypes. b Relationship of durable clinical benefit with TICs infiltration (DCB, n = 14; NDB, n = 12). P values were calculated using the two-sided Mann-Whitney U test. The line in the middle of the box represents the median. The lower and the upper edges of the box are the 1st and 3rd quartiles, respectively. The dots are considered outliers, which are more than 1.5*IQR beyond the lower or upper quartiles
Association between molecular features and tumor-infiltrating cells
Considering the potential predictive significance of LRP1B and PI3K pathway status, we examined the association between LRP1B status and cell infiltration in the TME. The presence of LRP1B mutation was associated with a reduction in the infiltration of CD8+ T cells and PD-1+CD8+ T cells compared to tumors that did not contain LRP1B mutation (supplementary Table 5). Additionally, alterations in the PI3K pathway were associated with reduced infiltration of CD3+, CD4+, and CD8+ T cells (supplementary Table 6).
Association between spatial distribution characteristics and durable clinical benefit
We further analyzed the spatial distribution characteristics of TICs in AFP-G/GEJ adenocarcinoma. Lymphocytes including CD3+ T, CD4+ T, CD8+ T cells, and CD20+ B cells were mainly distributed in the stroma region, but myeloid cells such as CD66+ neutrophils, CD68+ macrophages were located more in the tumor parenchyma, highlighting the particular distribution of tumor-infiltrating immune cells in AFP-G/GEJ adenocarcinoma (supplementary Fig. 10). We then evaluated the prognostic value of the density of TICs by distribution region. The data indicated that CD8+ T cells and PD-1+CD8+ T cells exhibited a similar trend in both contexts. Specifically, the densities in the DCB were significantly higher than in the NDB (Supplementary Tables 7–8). With the precise localization of central cells and surrounding cells established, we then assessed the clinical relevance of the proximity between individual tumor cells (central cells) and TICs (surrounding cells) within the tumor parenchyma by the “effective score” parameter established previously.25 The radius (10μm) was preselected in order to identify surrounding cell populations that were likely capable of effective cell-to-cell interaction with central cells (supplementary Fig. 11). The effective scores of CD8+ T cells, PD-1+CD8+ T cells, FOXP3+CD4+ T cells and FAP+ cells, which represents cancer-associated fibroblasts (CAFs), were higher in the DCB than those in the NDB (Fig. 5a, b). Also, as our treatment regimen included the anti-angiogenic agent apatinib, we further investigated the effective scores of the cells surrounding the CD31+ cells within 10 μm radius to explore the interactions between TICs and blood vessels. Notably, the effective scores of CD3+ T cells and FAP+ cells were higher in the DCB group, whereas the effective scores of CD8+ T cells did not differ between the two groups (Fig. 5c, d; supplementary Fig. 12). Recent studies suggest that perivascular CAFs chemotactically recruit T cells into the tumor parenchyma.26 In this study, we observed a positive correlation between the effective score of CAFs around tumor blood vessels and the effective score of T cells. In contrast, no such correlation was observed for CAFs surrounding tumor cells (supplementary Fig. 13a, b).
Spatial analysis of AFP-G/GEJ adenocarcinoma shows a hierarchy of organization of TICs in diffirent treating efficacy. a Representative multiplex immunofluorescence images of spatial analysis from DCB group and NDB group. Arrowheads in left rows denote CD8+ T and PD-1+CD8+ T cells while those in the right denote FOXP3+CD4+ T cells. b The distribution of the effective score of TICs populations in the tumor parenchyma in 10 µm increments between DCB (n = 14) and NDB (n = 12). Error bars represent mean ± SEM. c Representative multiplex immunofluorescence images of spatial analysis from DCB group and NDB group. Arrowheads denote CD3+ T cells and CD31+ cells. d The effective score of CD3+ T cell populations around CD31+ cells in the tumor parenchyma in 10 µm increments between DCB (n = 14) and NDB (n = 12). P values in b and d were calculated using the two-sided Mann-Whitney U-test. The line in the middle of the box represents the median. The lower and the upper edges of the box are the 1st and 3rd quartiles, respectively. The dots are considered outliers, which are more than 1.5*IQR beyond the lower or upper quartiles
Increased density of T cells upon treatment
Finally, we assessed the temporal dynamics of TME corresponding to the ICIs-combined treatment. Post-treatment, a significant increase in both CD3+ T cells and CD8+ T cells was observed, suggesting that camrelizumab plus apatinib and SOX could recruit T cells and remodel a more ‘inflamed’ TME in AFP-G/GEJ adenocarcinoma (supplementary Fig. 14).
